Selectivity and Promiscuity of the First and Second PDZ Domains of PSD-95 and Synapse-associated Protein 102*

PDZ domains typically interact with the very carboxyl terminus of their binding partners. Type 1 PDZ domains usually require valine, leucine, or isoleucine at the very COOH-terminal (P 0 ) position, and serine or threonine 2 residues upstream at P (cid:1) 2 . We quantitatively defined the contributions of carboxyl-terminal residues to binding selectivity of the prototypic interactions of the PDZ domains of postsynaptic density protein 95 (PSD-95) and its homolog synapse-associated protein 90 (SAP102) with the NR2b subunit of the N -methyl- D -aspartate-type glutamate receptor. Our studies indicate that all of the last five residues of NR2b contribute to the binding selectivity. Prominent were a requirement for glutamate or glutamine at P (cid:1) 3 and for valine at P 0 for high affinity binding and a preference for threonine over serine at P (cid:1) 2 , in the context of the last 11 residues of the NR2b COOH terminus. This analysis predicts a COOH-termi-nal (E/Q)(S/T) X V consensus sequence for the strongest binding to the first two PDZ domains of PSD-95 and SAP102. A search of the human genome sequences for proteins with a COOH-terminal (E/Q)(S/T) X V motif yielded 50 proteins, many of which have not been previ-ously identified as PSD-95 or SAP102 selectivity PDZ PSD-95 constants measuring FP of fixed concentrations of fluo-rescein-labeled peptides increasing concentrations of PDZ Sequence selectivity by these in-solution measurements corroborated by an ELISA-styled assay with biotinylated peptides, which were attached to streptavi-din-coated plates and with increasing concentrations of PDZ domains.

PDZ domains are unique among protein-protein interaction domains, because the proteins that bind to PDZ domains generally do so by their very COOH-terminal residues. The subunits of the Shaker K ϩ channel and NMDA receptor that bind to the first two PDZ domains of PSD-95, SAP102, and PSD-93 have the sequence E(S/T)DV-COOH. The co-crystal structure of the third PDZ domain of PSD-95 and a peptide ligand derived from the very COOH terminus of CRIPT (TKNYKQTSV-COOH) has been solved (23). This work showed that the very COOH-terminal valine and the threonine two positions upstream (the 0-and Ϫ2-positions P 0 and P Ϫ2 , respectively) form crucial interactions in the PDZ domain binding pocket. Thus, the minimal consensus sequence for binding to the PSD-95 PDZ domains has been defined as an (S/T)XV motif, where X can represent any residue. However, it is clear that other residues must contribute to specificity for a given PDZ domain, because various proteins and ion channels that have an (S/ T)XV motif do not bind any of the PDZ domains of PSD-95 under conditions under which the other ligands do. Examples include the neuronal inwardly rectifying K ϩ channels Kir3.2 and Kir3.3 (COOH-terminal sequence is in both cases ESKV (24)); the Na ϩ channel Na v 1.5 (ESIV (25)), which is present not only in muscle but also brain (26); and diacylglycerol kinase (ETAV (27)). Furthermore, the ␤ 1 adrenergic receptor does not interact with the first two PDZ domains of PSD-95 although it does conform to the (S/T)XV motif (SKV (28)). This receptor does, however, bind to the third PDZ domain of PSD-95 (28). This PDZ domain possesses a binding preference that is quite different from the first two PDZ domains (e.g. see Ref. 12) of PSD-95 and does not interact with NR2 subunits (9,29) or Shaker-type K ϩ channels (10). Similarly, Neuroligin carries the sequence TRV at its COOH terminus but only interacts with the third and not the first two PDZ domains of PSD-95 (30).
PDZ domains can be broadly divided into several categories, based on their general ligand specificity. Type I PDZ domains, found on PSD-95 and its homologs, bind (S/T)X(V/I/L) COOH termini, whereas type II PDZ domains (e.g. the fourth and fifth PDZ domain of GRIP and the PDZ domain of CASK), bind an ⌽-X-⌽ motif, where ⌽ represents a hydrophobic residue (preferably tyrosine or phenylalanine at P Ϫ2 ) (31,32). A third type of PDZ domain present in neuronal nitric-oxide synthase shows a preference for aspartate at P Ϫ2 (DXV motif) (33,34), although it also accepts other residues at P Ϫ2 (e.g. isoleucine (35)). Additionally, another kind of binding has been described and is exemplified by an internal (non COOH-terminal) sequence in neuronal nitric-oxide synthase that binds to syntrophin's PDZ domain and the second PDZ domain of PSD-95 (7,36).
The interaction between the NMDA receptor NR2b subunits and PSD-95 is one of the best defined PDZ domain interactions. The NMDA receptor is critical for a number of neuronal functions such as synaptic transmission and synaptic plasticity. PSD-95 and its homologs may be crucial for the assembly of postsynaptic proteins and signal transduction involving the NMDA receptor (8,13,37,38). Therefore, we evaluated the role of the various residues at the very COOH terminus of the NMDA receptor in determining the affinity and specificity of its interaction with PSD-95 and its relative, SAP102.
Various combinatorial methods have been employed to study the specificity of PDZ domains. The yeast two-hybrid method has been used extensively in the characterization of PDZ domains and their selectivity. But although it is a powerful and sensitive assay, it is prone to give a high background of false positives and is only useful for a qualitative assessment of binding. Precise affinity determinations are not possible. Oriented peptide libraries have been used to determine the consensus binding sequence for a variety of PDZ domains (32). Different PDZ domains were immobilized and incubated with batches of solubilized peptide libraries. Peptides retained by the PDZ domains were sequenced, and the frequency of residues at each position was quantified. An increase in the probability for a given residue at a certain position by a factor of 1.5 was considered to be significant. This method can reveal preferences of the various PDZ domains for certain residues at a given position but does not allow quantitative comparisons of affinities of defined peptides. Because this method does not determine actual sequences of any of the PDZ domain-binding peptides, it does not permit judgment of whether a preference for a residue at one position is linked to the presence of another residue at another position. Furthermore, the method used by Songyang et al. (32) is best suited for detecting preferences critical at a given position for high affinity binding. However, it would be difficult if not impossible to determine with this method whether certain residues are forbidden at a given position if this position accepts most of the other residues. Most recently, a COOH-terminal phage display has been used to explore the specificity of the seven PDZ domains of the human homologue of InaD (39). This approach identified putative consensus binding sequences for each PDZ domain, although many interactions may have been of low affinity, perhaps with K D values above 10 M, because the high display density on the phage surface may have resulted in avidity effects (39). As for the other methods, it is not possible to determine affinities of the phage-displayed peptides and compare the precise contribution of each residue.
To understand how each position in the PDZ binding motif may contribute to the affinity for the PDZ domain, we used several different libraries of selected peptides based on the COOH terminus of the NR2b subunit of the NMDA receptor to identify the important positions and residues involved in the binding of NR2b to the PDZ domains of PSD-95 and SAP102. The use of solid-phase colorimetric and in-solution fluorescence polarimetric peptide binding assays allowed us to quantitatively determine the effects of amino acid substitutions at various positions along the peptide sequence on PDZ domain binding. We were surprised by a strong, often severalfold preference of the first two PDZ domains of PSD-95 and of SAP102 for valine at the 0-position over the related amino acids leucine and isoleucine, for threonine at the Ϫ2-position over serine, and for glutamate and glutamine at the Ϫ3-position over aspartate. Analysis of the human genome indicated that besides NMDA receptor NR2 subunits and K ϩ channels of the Shakertype Kv1 family, only a limited number of other neuronal proteins possess COOH termini that would predict a strong interaction with the first two PDZ domains of PSD-95 or SAP102. Despite the large number of proteins with COOHterminal (S/T)XV motifs that can potentially bind to PDZ domains, our analysis indicates a clear selectivity of specific PDZ domains for a limited number of proteins.

EXPERIMENTAL PROCEDURES
Production of GST Fusion Proteins-NH 2 -terminal GST fusion proteins of the second PDZ domain of rat PSD-95 (residues 154 -248) and those carrying residues 142-235 (PDZ1), 237-333 (PDZ2), and 399 -485 (PDZ3) of rat SAP102 were expressed from a modified pGEX2T vector named pGHEB and kindly provided by Dr. Craig C. Garner (University of Alabama, Birmingham, AL). To create vectors for the expression of GST fusion proteins of the first and third PDZ domain of PSD-95, cDNA sequences encoding human PSD-95 residues 82-202 (PDZ1) and 344 -443 (PDZ3; residue numbers refer to full-length human PSD-95 sequence as given in Ref. 40) were excised with EcoRI from corresponding GAD10 vectors, which contained the respective sequences that had originally been inserted into the GAD10 EcoRI cloning sites (10) (kindly provided by Morgan Sheng, MIT, Cambridge, MA). The latter two protein sequences are completely identical to residues 39 -159 and 301-400 encoding the first and third PDZ domain of rat PSD-95, respectively. DNA fragments were purified by agarose gel electrophoresis and ligated into EcoRI-digested pGEX 4T-1 (Amersham Biosciences) before screening the resulting clones for correct orientation of the inserts. All vectors were confirmed by DNA sequencing with the Ampli-Taq system (PerkinElmer Life Sciences). Fusion proteins were expressed and purified as described (41,42) with the following modifications. The vector DNA was electroporated into the E. coli strain BL21 -DE3 cells. 50-ml cultures were grown overnight, diluted 1:10 with LB medium, and incubated until the cultures reached an A 600 of 1.0 (2-4 h). Fusion protein expression was induced with isopropyl-␤-Dthiogalactopyranoside (100 M, 2 h). Cells were harvested; resuspended in TBS (10 mM Tris-Cl, pH 7.4, 150 mM NaCl); digested with lysozyme; supplemented with 10 mM EDTA, 15 mM dithiothreitol, and protease inhibitors (0.2 mM phenylmethanesulfonyl fluoride, 1 g/ml pepstatin A, 0.1 g/ml leupeptin, 0.1 g/ml aprotinin); and lysed by adding 1.5% (final concentration) sarkosyl (43), followed by sonication. Lysates were clarified by ultracentrifugation (45-Ti rotor, 40,000 rpm, 186,000 ϫ g, 1 h). Sarkosyl was neutralized by adding 2-4% (final concentration) Triton X-100 from a 20% stock solution. Lysates were incubated with glutathione-Sepharose (Amersham Biosciences) overnight at 4°C. Resins were washed with TBS and eluted by adding 15 mM glutathione in 150 mM NaCl, 50 mM Tris-Cl, pH 8. The fusion proteins were dialyzed against TBS and subsequently quantified by determining A 280 and, in parallel, by the bicinchoninic acid assay (Pierce) and a Coomassie assay (Pierce) in microtiter plate format. The purity and quality of all fusion proteins was evaluated by SDS-polyacrylamide gel electrophoresis and detection by staining with Coomassie Brilliant Blue and by immunoblotting with anti-GST antibodies to ensure minimal degradation. Purified GST fusion proteins usually did not show contamination with other proteins by Coomassie Brilliant Blue staining and were discarded otherwise.
Peptide Synthesis-Peptides based on the NR2b COOH terminus and the COOH termini of other potential PDZ ligands were synthesized manually using a Multipin Peptide Synthesis kit according to the manufacturer's instructions (Chiron-Mimotopes) (44). This method utilizes standard solid phase peptide synthesis with Fmoc-protected amino acids, followed by deprotection and cleavage of the peptide from the solid-phase support with trifluoroacetic acid. We typically synthesized peptides using a 96-well plate array of polystyrene pins serving as the solid phase support. After cleavage, peptides were cleaned with two ether/petroleum ether extractions to remove leftover protecting groups.
The remaining trifluoroacetic acid was eliminated by dissolving the peptide in water/acetic acid/acetonitrile followed by lyophilization. Peptides were dissolved in water and stored frozen until used. All peptides contained at their NH 2 termini a lysine, followed by a serine-glycine spacer. The NH 2 -terminal lysine residue was modified with a biotin or fluorescein tag (Fmoc-biotinyl-lysine and Fmoc-fluoresceinyl-lysine were from Anaspec). The NH 2 -terminal tag-KSG sequences were followed by 11 residues corresponding to the wild type (WT) or modified COOH termini of NR2b and of other PDZ-binding proteins. Selected peptides were analyzed by mass spectrometry, and all of the peptides were quantified with a BCA assay.
Fluorescence Anisotropy Plate Assays-Saturation binding curves for PDZ domain-peptide interactions were determined by examining the change in fluorescence polarization (FP) (45) of a fixed concentration of the various fluorescein-labeled peptides (100 nM) with increasing amounts of a GST-PDZ domain fusion protein. Peptides and fusion proteins were diluted using TBS with 1% bovine serum albumin to prevent absorption of the peptide to the plate. After combining the fluorescent peptide and PDZ domain in a black 384-well plate and allowing the system to come to equilibrium (1-2 h), the plate was read on a Victor 2 V (PerkinElmer Life Sciences) plate reader in the fluorescence polarimeter mode to obtain polarization values (P). P is calculated according to the equation P ϭ (I v Ϫ g * I h )/(I v ϩ g * I h ), with I v and I h corresponding to the vertical and horizontal fluorescence intensities, respectively; g is a calibration correction, obtained from reading the polarization value of a 100 nM fluorescein solution, which possesses a P value of 0.027 under our conditions at room temperature. The g value is calculated from the equation g ϭ (I v0 /I h0 ) * (1 Ϫ 0.027/1 ϩ 0.027), where I v0 and I h0 are the vertical and horizontal fluorescent intensities of the fluorescein solution in TBS plus 1% bovine serum albumin. The P values of fluorescent peptide solutions without PDZ domains were subtracted from the P values obtained by titrations with PDZ domains. The resulting P values were plotted against the concentration of the PDZ domain. To determine the K D value, a curve was fitted by the equation Y ϭ B * X/(K D ϩ X), with B being the maximum P value that would be reached at saturation as indicated by the extrapolation of the fitted curve.
Solid-phase Colorimetric Plate Assays-Streptavidin-coated 96-well microtiter plates (Pierce) were incubated with 1% bovine serum albumin in TBS and subsequently with saturating amounts of biotinylated peptides. After two washing steps with TBS to remove any unbound peptide, GST-PDZ domain fusion proteins were added in TBS at varying concentrations and allowed to incubate for 1-2 h. Following two quick washing steps with TBS, bound GST fusion proteins were detected by incubation with anti-GST antibodies (41) for 1 ⁄2 h, washed twice with 0.1% Triton X-100 in TBS, and incubated with Protein A coupled to horseradish peroxidase (HRP; 1:10,000 in TBS) for 1 ⁄2 h. After washing with 0.1% Triton X-100 in TBS, 100 l of the colorimetric HRP substrate solution "Slow-TMB" (Pierce) was added to each well. After 5 min, the reaction was stopped with 100 l of 1 M H 2 SO 4 , and the plate was read at 650 nm.

RESULTS
Affinity Determinations of Fluorescein-labeled Peptides for the First Two PDZ Domains of PSD-95 and SAP102-Type I PDZ domains typically bind to SXV consensus sequences located at the very COOH termini of their interaction partners. Bulk sequencing of peptides that bound to PDZ domains during incubation with peptide libraries indicated some preference (in the range of 1.5-2-fold) for certain amino acids over others at most positions within the COOH-terminal 10 residues (32). However, residues of the COOH-terminal (S/T)XV motif for a given PDZ domain can be quite flexible. For example, serine at the Ϫ2-position (P Ϫ2 ) can usually be substituted by threonine as is the case for Kv1.4 (10), and isoleucine or leucine can often replace valine at P 0 (51-55). Our goal was to quantify sequence selectivity of the first two PDZ domains of PSD-95 and SAP102 for target binding using defined peptides. We determined binding constants by measuring FP of fixed concentrations of fluorescein-labeled peptides with increasing concentrations of PDZ domains. Sequence selectivity indicated by these in-solution measurements was corroborated by an ELISA-styled assay with biotinylated peptides, which were attached to streptavidin-coated plates and incubated with increasing concentrations of PDZ domains.
Constant concentrations of the NR2b WT peptide were incubated with increasing concentrations of purified GST-PDZ fusion proteins, and FP was determined with a Victor 2 V plate reader (Fig. 1). The peptide contained the 11 COOH-terminal residues of NR2b and the linker sequence KSG at the NH 2 terminus with the NH 2 -terminal lysine fluorescein tagged. The rotational mobility of this fluorescent peptide decreased when bound to a PDZ fusion protein. Therefore, as the proportion of fluorescent peptide bound to PDZ fusion proteins increased, the aggregate FP of the solution increased. The measured values could be fitted to saturation curves from which the K D values were determined. We found that the second PDZ domain of PSD-95 and SAP102 bound with higher affinity than the first PDZ domain (0.91 versus 2.3 M for PSD-95; 0.64 versus 1.41 M for SAP102). These values are in good agreement with those obtained earlier for in-solution interaction assays for PDZ domain binding to their respective ligands (12,56). It is also interesting to note that SAP102 had slightly higher affinity for the NR2b WT peptide than PSD-95 for corresponding PDZ domains.
Substantially lower K D values have been observed by solid phase-based assays including surface plasmon resonance measurements (30,57). However, these solid phase measurements were based on interactions of immobilized peptides carrying respective COOH-terminal SXV motifs with recombinant proteins that contained more than one PDZ domain. Accordingly, these solid phase assays may not reflect the true value for a single PDZ domain-target interaction because one protein may simultaneously bind two or more of the immobilized peptides, thereby dramatically reducing its off-rate and causing rebinding and avidity effects (58,59). Similar considerations apply for experiments with single PDZ domains of SAP102 expressed as GST fusion proteins (4) because the GST moieties can dimerize and thereby result in avidity effects in solid phase assays (58). To avoid these problems, we primarily worked with peptides in solution. In our solid phase assays, we used GST fusions with a single PDZ domain, and the wells were coated with our peptides at low density.
We tested peptides derived from proteins that were initially identified as binding partners for one or more of the three PDZ domains of PSD-95 and were either strong or weak interactors ( Fig. 2A). Highest affinities were observed for binding of the second PDZ domains of PSD-95 and SAP102 to NR2a, NR2b, NR2c, and the K ϩ channel Kv1.4. Binding to the Kv1.1 K ϩ channel and the COOH terminus of the NR1 C2Ј domain, which constitutes one of the two alternatively spliced COOH termini of the NR1, was much weaker. The first PDZ domains of PSD-95 and SAP102 exhibited substantially lower affinities for these peptides than the second PDZ domains but showed a similar ranking order, with NR2a, NR2b, NR2c, and Kv1.4 being much stronger interactors than Kv1.1 and NR1 C2Ј. The relative ranking of all of these affinities agrees with earlier semiquantitative data. For example, yeast two-hybrid assays suggested much stronger interactions of the second than of the first PDZ domain of PSD-95 with NR2a, NR2b, and NR2c (29), and PSD-95 interacts much better with Kv1.4 than with Kv1.1 in this assay (10). NR1 is generally not considered to bind to PSD-95 (60). However, the C2Ј COOH terminus does conform  . NR2b⌬4 refers to a peptide that was NH 2 -terminally shifted by 4 positions away from the COOH terminus, eliminating the COOH-terminal ESDV sequence. FP measurements were performed, and K D values were calculated as described under "Experimental Procedures." To facilitate reading of the relative binding affinities, K D values were normalized to the WT NR2b K D , which was set to equal 1, and the inverses of the normalized values were plotted as bars. Numbers beside the bars give the actual K D values (in M; N.D., not determined). The peptide sequence is listed in the far right column. K D values greater than the 10-fold K D for WT NR2b are generally listed in the chart as 0, because accurate values for peptides with weak affinity were difficult to determine. Each assay was usually done in quadruplicate.
to the (S/T)XV motif. Several pieces of evidence suggest that the NR1 C2Ј region can associate with PSD-95 and SAP102, although it is unknown whether this interaction is mediated by the first two or rather the third PDZ domains of these two proteins (9,61,62). The low affinities observed for the Kv1.1 and NR1 C2Ј-derived peptides may therefore reflect that binding of C2Ј to the first two PDZ domains is not as firm as for the other interactions described above and may require addition interactions for stabilization. PSD-95 and SAP102 may form interactions to NR1 in parallel to those with NR2 subunits in the NMDA receptor complex, thereby increasing the stability of the complex. However, we cannot rule out the possibility that Kv1.1 and NR1 does not at all associate with one of the first two PDZ domains of PSD-95 or SAP102 in vivo. Removal of the last 4 residues of NR2b by shifting the peptide sequence 4 positions toward the NH 2 terminus exposes another sequence at the very COOH terminus that constitutes an (S/T)XV motif with isoleucine substituting for valine at the COOH terminus. However, this peptide did not show detectable binding at all.
We also analyzed the interactions to the first two PDZ domains of PSD-95 and SAP102 with the COOH termini of Neuroligin and CRIPT, two proteins that bind to the third PDZ domain of PSD-95. As expected (12,30), the Neuroligin peptide did not show any detectable binding; however, the CRIPT peptide exhibited relatively strong binding to the second, although not the first, PDZ domains of PSD-95 and SAP102. In fact, the COOH-terminal CRIPT sequence comes close to the optimal binding sequence for the first two PDZ domains of PDS-95 and SAP102 (see below), and the original work on CRIPT described a weak interaction of this protein with the second, although not the first, PDZ domain of PSD-95 in the yeast two-hybrid system (12). In summary, the results described in this and the previous paragraphs correspond well to earlier qualitative and semiquantitative observations and afford a reliable survey of binding constants. Accordingly, our assay is well suited to determine the contribution of certain residues at the COOH terminus of NR2b for PDZ domain binding by comparing the resulting affinity values of the various peptides with alterations in their sequences.
The COOH termini of GluR1 (TGL) and GluR2 (VKI) subunits of the AMPA-type glutamate receptors constitute consensus sequences for binding type I and type II PDZ domains, respectively. GluR1 has been shown to be associated with SAP97, which, like PSD-95, contains type I PDZ domains, but no interaction with PSD-95 or SAP102 was detectable (41,50). GluR2 binds to type II PDZ domains present in GRIP1 and -2 (19,20). Peptides derived from these COOH termini did not show detectable binding (Fig. 2B), further confirming the specificity of our assays. We produced chimeric peptides based on the GluR1 COOH-terminal sequence. Starting at the very COOH-terminal position P 0 , we replaced an increasing number of GluR1 residues with those of NR2b. Little specific binding was detectable until at least the last 4 GluR1 residues (i.e. P Ϫ3 -P 0 ) were replaced with those from NR2b (Fig. 2B). Substituting positions P Ϫ4 and P Ϫ5 further improved binding of the peptides, especially to the PDZ1 domains, which required these substitutions for detectable peptide interactions. Of note, binding by PDZ2 of PSD-95 and SAP102 was not very sensitive to substituting isoleucine for glycine at P Ϫ4 in these experiments, in contrast to those with single point mutations described below. This difference may reflect that contributions of each position may be influenced by the context of the other positions, including those upstream of P Ϫ4 , although these latter positions appear to have little influence on binding when probed by single point mutations (see below).
Contribution of the Residues of the Very COOH Terminus of NR2b to Binding to the First Two PDZ Domains of PSD-95 and SAP102-To evaluate the role of each position at the COOH terminus of NR2b in PDZ binding, we systematically introduced single point mutations in this sequence and determined the K D values of the resulting peptides by FP (Fig. 3). We also evaluated the relative contributions of each position in solid phase assays with the corresponding biotinylated peptides anchored at streptavidin plates. The density of streptavidin in the plate wells was low, a characteristic of the plates we used that helped to avoid avidity and rebinding effects. For most biotinylated peptides, the solid phase binding assays were performed at single concentrations of the PDZ fusion proteins (Fig.  4), but for selected peptides, titration curves with increasing amounts of the PDZ fusion proteins were obtained (Fig. 5). These solid phase assays corroborated the relative contribution of each residue to PDZ binding (compare Fig. 3 with Figs. 4 and 5). The apparent K D values calculated from the solid phase titration assays were generally higher than those for the FP measurements, usually by a factor of 2-4. This increase in the . Each peptide carried a single substitution at one of the 11 COOH-terminal positions designated P 0 (the very COOH-terminal position) to P Ϫ10 . The column on the left indicates the residue in the WT NR2b sequence, followed by the position number and the amino acid residue substituted. Numbers following each bar indicate measured K D values, whereas bars represent inverse K D values normalized to WT NR2b, which equals 1, to facilitate survey of relative binding affinities. K D values greater than the 10-fold K D for WT NR2b are generally listed in the chart as 0, because accurate values for peptides with weak affinity were difficult to determine. Each assay was typically performed in quadruplicate. apparent binding constant is due to unbinding of ligand during the incubation and washing steps subsequent to the initial binding reaction of the purified PDZ fusion proteins with the immobilized peptides. Those steps are performed after the removal of unbound PDZ domains and take a minimum of 1.5 h. If longer washing periods are applied, the apparent K D values increase further (data not shown). Therefore, the solid phase titration assays provide not true but only "apparent" K D values. However, the ranking order from these solid-phase titration assays agrees very well with that obtained by FP and therefore corroborates the latter.
Initial experiments on the interaction of PSD-95 with NR2 and Kv1.4 COOH termini, which were the first studies identifying PDZ domain interactions with target proteins, indicated that the serine (NR2 subunits) or threonine (Kv1.4) at P Ϫ2 and the valine at P 0 are the most crucial residues and cannot effectively be replaced by alanine, hence the reference to the "(S/T)XV" motif. Later studies on other type I and also type II PDZ domains indicated that valine at P 0 can be replaced by other hydrophobic residues including leucine and isoleucine (51)(52)(53)(54)(55). We observed that valine at P 0 was by far superior for binding to the first two PDZ domains of PSD-95 and SAP102 as compared with other hydrophobic residues, including leucine, isoleucine, and phenylalanine. The presence of valine versus leucine or isoleucine at P 0 may therefore be crucial in determining the specificity of a COOH-terminal interaction with a PDZ domain, with valine increasing the selectivity for PSD-95 and SAP102.
NR2 subunits, Kv1.1, and Kv1.4, carry either aspartate or glutamate at P-1. Substituting glutamate for aspartate in the NR2b peptide sequence did not cause big changes in affinities for PDZ1 or -2 of PSD-95 and SAP102. Glutamine-substituted peptides bound equally well, and so did asparagine-substituted ones, at least for PDZ domain 95/1 in the solid phase single concentration survey assay (Fig. 4). However, asparagine substitution substantially reduced binding to the other three PDZ domains in that assay (we did not perform FP assays for this substitution). Serine showed a profile similar to that of asparagine, with nearly unaltered binding affinity for PDZ domain 95/1 but reduced affinity for the other three PDZ domains. Alanine-substituted peptides showed good binding, although somewhat weaker than that of WT, with affinities in the range around 2-4 M. A change to glycine or lysine strongly reduces binding affinities with K D values above 10 M, a concentration range at which the precise K D values often proved hard to determine with our survey assays. An exception is PDZ domain 102/2, which exhibited a K D value around 4 M for the glycinesubstituted peptide, suggesting that glycine can be tolerated at this position to some degree, at least by certain PDZ domains. In summary, P Ϫ1 is quite flexible, and our PDZ domain interactions tolerate very well various residues including aspartate, glutamate, glutamine, and alanine; asparagine and serine work well for 95/1 binding, and glycine works to some extent for 102/2 interaction, although binding to other PDZ domains appears compromised for peptides with the last three substitutions. The positively charged residue lysine is generally not acceptable at P Ϫ1 for binding to these PDZ domains.
Structural studies indicate that either a serine or a threonine is required at P Ϫ2 for type I PDZ domain interactions; serine and threonine at this position form a hydrogen bridge with the histidine residue at the first position of the ␣B helix (see Fig. 8), which is conserved in type I PDZ domains (see Table III). The Ϫ2-position of NR2b and of other NR2 subunits has a serine. Replacing this serine with threonine as present in Kv1.4 and Kv1.1 resulted in a substantial increase in binding affinities for all four PDZ domains (Figs. 3 and 5). As expected, an alanine replacement caused a strong decrease in binding to most PDZ domains, although some binding in the low micromolar range was observed for 102/2 ( Fig. 3; see also Fig. 4). Peptides with the bulky tryptophan at this position did not show binding (Fig. 4). Accordingly, serine at P Ϫ2 can be replaced by threonine, which increases binding, but other residues are not well tolerated.
Glutamate is conserved at the Ϫ3-position in NR2 subunits and Kv1.4, but this position was initially considered less critical for PDZ binding, in part because Kv1.1, although a weaker interactor than Kv1.4 (10) (see also Fig. 2), carries a hydrophobic residue at this position and because substituting glutamine for glutamate in Kv1.4 did not alter its interaction with PSD-95 (9, 10). We found that only glutamate or glutamine at P Ϫ3 allows for high affinity interaction with PDZ1 and -2 of PSD-95 and SAP102. Other substitutions including those with aspartate, serine, threonine, alanine, leucine, phenylalanine, or lysine substantially reduced or eliminated binding. The fact that aspartate but not glutamine substitutions decreased binding indicates that the length of the side chain but not the negative charge is critical at this position for interaction with our PDZ domains. Both the -carboxyl group of the glutamate residue and the amide group at the -position of the glutamine residue are capable of forming hydrogen bonds, and this ability seems to be important (see below).
Like the Ϫ1-position, the Ϫ4-position appears to tolerate a variety of residues. NR2a and -2b, NR2c and Kv1.1, and Kv1.4 and NR1 have isoleucine, leucine or valine, respectively, at this position, and replacing the valine in Kv1.4 with either arginine or tryptophan did not alter PSD-95 binding in yeast two-hybrid interaction assays (10). In addition, more recently identified binding partners for PDZ1 and -2 of PSD-95 include Kir2.1 and -2.3 (24,51), and ErbB4 (63), which possess an arginine at that position. Our peptide binding assays confirm that P Ϫ4 of NR2b is quite flexible and accepts not only hydrophobic residues (e.g. leucine; Figs. 4 and 5) but also lysine, which carries a positive charge similar to arginine. However, glycine substitution resulted in a strong loss of binding, indicating that the removal of a side chain does disturb binding. This position is, therefore, not completely neutral. In fact, alanine, methionine, and aspartate substitutions also strongly reduce binding. Collectively, these data indicate that P Ϫ4 accepts various hydrophobic and positively charged residues, but the presence of other residues including glycine, alanine, methionine, or aspartate selects against binding to PDZ1 and -2 of PSD-95 and SAP102.
Positions Ϫ5 through Ϫ10 generally accepted most substitutions, suggesting that they are less critical in determining PDZ domain interactions. However, aspartate was not well tolerated at P Ϫ5 (Figs. 3 and 4). Replacing tyrosine with an aspartate at P Ϫ10 resulted in a strong increase in binding affinity for the second PDZ domains of PSD-95 and SAP102 in the FP assay ( Fig. 3) but not the solid phase assay (Fig. 4). It is quite possible that this position is too close to the plate surface to be fully accessible in the solid phase assay for interaction with the PDZ domain and may, therefore, only show an effect in the FP assay. These results suggest that selected positions past the P Ϫ4 may make some, although quite limited, contributions to PDZ domain interactions.
Affinities of Peptides for the Third PDZ Domains of PSD-95 and SAP102-We also tested the CRIPT-and Neuroligin-derived peptides along with those corresponding to the COOH termini of the various glutamate receptor subunits and Kv1.4 in binding assays with the third PDZ domains of PSD-95 and SAP102. The CRIPT peptide exhibited by far the highest affinity for both PDZ domains with a K D in the range of 3 M for PDZ3 of PSD-95 as determined by titration of FP (Fig. 6A). Neuroligin binding was also saturable but showed a much higher K D of 28 M in this assay (Fig. 6A), whereas the NR2b wild-type peptide does not show specific binding for PDZ3 (Fig.  6B). For nearly all of the peptides based on the NR2b COOHterminal sequence, FP and the solid phase assay indicated little to no specific binding (data not shown). However, the NR2c peptide consistently showed distinguishable binding above background (Fig. 6B). In fact, FP titration indicated that association of this peptide with PDZ3 is saturable with a K D around 25 M (Fig. 6A). This finding opens up the possibility that the COOH terminus of NR2c may not only interact with the first two but also the third PDZ domains of PSD-95 and SAP102. Although the affinity is relatively low for the latter interactions, they may help stabilize association of NR2c with the first two PDZ domains of PSD-95 and SAP102. In a similar way, CRIPT may not only interact with the third but also with the second and perhaps first PDZ domains of PSD-95 and SAP102. The idea of combining weak and strong COOH-terminal interactions with PDZ domains also receives support by early findings that NR2 subunits and Kv1.4 interact most robustly with the second PDZ domains of PSD-95 and SAP102 and more weakly with the first PDZ domains (10,29). Because each NMDA receptor and each Kv1.4 channel possesses several subunits with appropriate COOH-terminal SXV motifs, simultaneous interactions of one receptor or channel complex with several PDZ domains appears likely.
Survey of the Available Human Genome Sequences for Potential Binding Partners for the First Two PDZ Domains of PSD-95 and SAP102-To obtain an impression of how many potential binding partners with a COOH-terminal (E/Q)(S/ T)XV motif exist, we screened proteins as predicted by the currently available draft version of the human genome for this motif. We did not use any restrictions at the Ϫ1-position and Ϫ4-position. We identified 54 proteins and divided them into four categories (Table II). Group 1 contains all proteins with a COOH-terminal (E/Q)(S/T)(D/E/Q/N)V sequence. Of note, nearly all proteins in that group carry a hydrophobic or positively charged residue at P Ϫ4 , although this had not been a criterion for grouping those proteins together. We predict that these proteins are likely to bind strongly to the first two PDZ domains of PSD-95 and SAP102. In fact, five of the 14 members of this group had been identified as binding partners (in boldface type at the top of the list in Table II), and we present evidence that at least one member of the brain-specific angiogenesis inhibitor family (BAI1) does also interact with PSD-95 (see below). Group 2 proteins differ from group 1 by having a residue different from the binding-promoting aspartate, glutamate, asparagine, or glutamine at P Ϫ1 but carrying an advantageous aliphatic or positively charged residue at P Ϫ4 (i.e. isoleucine, leucine, valine, arginine, or lysine). Only two of the 16 proteins in this category are actually known to interact with the first two PDZ domains of PSD-95 or SAP102 (PMCA4b and CRIPT). Group 3 proteins have no favorable residue at either P Ϫ1 or P Ϫ4 , and only two of 20 proteins have been shown to bind to PSD-95 or SAP102. Finally, group 4 contains proteins with a lysine or an arginine at P Ϫ1 , which inhibit binding to our PDZ domains. Three of the four proteins, Kir3.2, Kir3.3 (24), and the ␤ 1 -adrenergic receptor (28), have actually been tested for binding to PDZ1 or -2 of PSD-95, with negative results. In the same studies, parallel pull-down experiments resulted in interactions that serve as positive controls (the inward-rectifying K ϩ channels Kir2.1 and Kir2.3 bound to PSD-95 in Ref. 24, presumably via the second PDZ domain (51), and the ␤ 1 adrenergic receptor bound to the third, although not the first or second, PDZ domain of PSD-95 in Ref. 28). These results confirm our finding that a lysine and perhaps an arginine are inhibiting binding to the first two PDZ domains of PSD-95 and SAP102.
A comparison with Table I, which lists all known interaction partners of the first two PDZ domains of PSD-95 and SAP102 with COOH-terminal SXV motifs, indicates that Table II identifies most proteins with COOH-terminal (E/Q)(S/T)XV sequences (NR2a to -d, Kv1.4, Kir3.2, 3.3, ␤ 1 adrenergic receptor, PMCA4b, CRIPT, Citron, and Sema4c in Table I). Maguin and SynGAP also possess COOH-terminal (E/Q)(S/T)XV sequences (Table I) but were not identified, because the sequences for these two proteins predicted from the human genome projects only show versions with truncated COOH termini, perhaps due to inaccuracies in the draft sequence (64). Kv1.1, Kv1.2, Kv1.3, Kir2.1, Kir2.3, ␤ 2 -adrenergic receptor, PMCA2b, ErbB4, Nedasin, and Neuroligin (Table I) do not conform to the (E/Q)(S/ T)XV motif and were not detected in our screen, which is, therefore, not complete in terms of binding partners for the first two PDZ domains of PSD-95 and SAP102. However, most of the latter proteins are known or predicted to interact with those PDZ domains more weakly than proteins with a COOH terminus that matches the (E/Q)(S/T)XV sequence (e.g. Kv1.4 and PMCA4b bind much stronger than Kv1.1-3 and PMCA2b, respectively; see "Discussion").
Not all of the proteins in Table II are actually potential binding partners of PSD-95 and SAP102 because their expression patterns do not overlap. For example, PSD-95 and SAP102 are only detectable in brain, whereas the PDZ binding kinase is abundant in testis and placenta and weakly expressed in heart but not brain (65,66). Tyrosinase-related protein 1 is involved in melanin syntheses and restricted to melanocytes. It binds to the PDZ domain of GIPC (67), but due to its restriction to melanocytes it would not constitute an interaction partner for PSD-95 or SAP102 in neurons. The Na ϩ channel Na v 1.5 was originally described as muscle-specific and interacts in these tissues with the PDZ domains of syntrophins (68). Recently, limited expression of Na v 1.5 in the brain has been described (26), but PDZ-domain interaction assays for a Na v 1.5-derived peptide were negative (25), suggesting that Na v 1.5 does not interact with PSD-95 or SAP102 even if co-expressed in some neurons. Another example of a protein that has a COOHterminal (E/Q)(S/T)XV motif and is present in neurons but does not appear to interact with PSD-95 is diacylglycerol kinase , which binds to the PDZ domains of syntrophins but not of PSD-95 (27).
Interaction of Brain-specific Angiogenesis Inhibitor BAI1 with PSD-95-Many proteins revealed by our proteomic search have not been considered in terms of association with PSD-95. We were able to obtain antibodies for three of these proteins (BAI1, PKC␣, and frizzled 2) and evaluated those proteins for co-immunoprecipitation with PSD-95 and SAP102 from rat brain extracts. Antibodies against BAI1 (48) recognize three bands in brain extract, all of which migrate in the range around 150 kDa; the lower two bands form a doublet, which is often not resolved ( Fig. 7A and data not shown). These bands are enriched after immunoprecipitation with the same antibodies (Fig. 7A), suggesting that all three polypeptides are specifically recognized by the antibody and represent various isoforms of BAI1. BAI1 immunoprecipitates also contained PSD-95, but SAP102 was never detectable in the BAI1 complex (Fig. 7A). To ensure that the BAI1-PSD-95 complex was present in the intact brain and did not form during homogenization, we homogenized parallel samples in a 50-fold larger buffer volume than under our standard conditions before collecting crude membranes by ultracentrifugation from both conditions. Subsequent immunoprecipitation resulted in comparable immunoreactivity for BAI1-associated PSD-95 (Fig. 7A). We did observe coprecipitation of BAI1 and PSD-95 and a lack of coprecipitation of BAI1 and SAP102 not only after extraction with 1% deoxycholate but also after extraction with either 1% Triton X-100 or 1% SDS (data not shown). Together with an analogous finding for PMCA2b, which also selectively binds to PSD-95 but not SAP102 (52), our observations that PSD-95, but not SAP102, is associated with BAI1 in the brain indicate that PSD-95 and SAP102 may play different roles in their cellular context by interacting with distinct, although partially overlapping protein pools. BAI1-related BAI2 and BAI3 also have a COOH-terminal SXV motif (YQTEV) with a hydrophobic residue at P Ϫ4 and a glutamate at P Ϫ1 and carry an acidic residue at P Ϫ10 . Because these residues at these positions increase binding of our NR2b-derived peptides to the first two PDZ domains of PSD-95, it is likely that BAI2 and -3 are also interaction partners for PSD-95.
In parallel with BAI1, we immunoprecipitated the Wnt receptor frizzled 2 (69 -71). The frizzled 2 antibody recognized one major band in brain extract of the expected molecular mass by immunoblotting (Fig. 7B). Immunoprecipitation with this

E/Q)(S/T)XV
We performed a BLAST search for COOH-terminal (E/Q)(S/T)XV-containing proteins. The search was restricted to the working draft sequence of the human genome protein products. For each match, the GenBank ™ accession number is given, along with the COOH-terminal consensus sequence (residues that potentially increase affinity for PSD-95 PDZ binding are in boldface type) and if there is evidence whether (ϩ) or not (Ϫ) the protein (or its mRNA) is expressed in the brain (? indicates that there is no evidence for brain expression, ϩ/Ϫ indicates that expression in the brain is low and spatially narrow). The entries are grouped (1 through 4) in order from the highest predicted affinity group to the lowest. Protein names in boldface type indicate confirmed association with PSD-95 PDZ domains.

Group 1. (E/Q)(S/T)(D/E/Q/
ϩ ? Hypothetical protein FLJ14050 R A R G P P R E S E V NP_060962 Ϫ PDZ-binding kinase; spermatogenesis-related protein

X(E/Q)(S/T)XV (X ‫؍‬ any except I/L/V/ K/R, X ‫؍‬ any except D/E/Q/N) NP_060259
ϩ Semaphorin 4c ϩ Rho GTPase-activating protein 6; rhoGAPX-1, isoform 1 and 4 D N P D A L P E T L V

NP_003042
ϩ Monocarboxylic acid transporter, member 1 ? Transforming growth factor, ␣ Ϫ Nucleosomal binding protein 1 D G K K E E P Q S I V NP_062550 Ϫ Class-I MHC-restricted T cell-associated molecule E K H I Q V P E S I V NP_000541 Ϫ Tyrosinase-related protein 1 E K L Q N P N Q S V V Group 4. (E/Q)(S/T)(K/R)V NP_002231 ϩ Inwardly rectifying K ϩ channel, Kir3.2 ϩ Inwardly rectifying K ϩ channel, Kir3.3 antibody, but not with a control antibody, resulted in the same band, indicating that this antibody effectively immunoprecipitated frizzled 2 under our conditions. However, we never observed coprecipitation of either PSD-95 or SAP102 with frizzled 2, whether membrane extracts were prepared with 1% deoxycholate, Triton X-100, or SDS ( Fig. 7A and data not shown). The COOH-terminal sequence of frizzled 2 matches that of frizzled 1 (GETTV) and is very similar to the COOH termini of frizzled 4 (SETVV) and frizzled 7 (GETAV). All of these sequences predict at best weak interactions with the first two PDZ domains of PSD-95 or SAP102 (group 3 in Table II). Our negative results suggest that frizzled 2 and its homologs do not bind to PSD-95 or SAP102 in vivo.
Interaction of PKC␣ with PSD-95 and SAP102-We performed similar experiments looking for co-immunoprecipitation of PKC␣ with PSD-95 and SAP102. PKC␣ is listed in group 2 (Table II) with an unfavorable alanine at P Ϫ1 but a beneficial leucine at P Ϫ4 in its COOH-terminal sequence (LQSAV), suggesting an intermediate affinity for the first two PDZ domains of PSD-95 and SAP102. PKC␣ has been shown earlier to bind with its COOH terminus to the PDZ domain of PICK1 (22). Both PSD-95 and SAP102 coprecipitated with PKC␣ (Fig. 7C). As in the previous experiments (Fig. 7A), immunoprecipitations with appropriate control antibodies were negative for PSD-95 and SAP102, indicating the specificity of the immunoprecipitations. Similar results were obtained when 1% Triton X-100 or 1% SDS instead of the routinely employed deoxycholate was used for solubilization of the membrane fractions (data not shown). Comparable amounts of PSD-95 and SAP102 coprecipitated with PKC␣ whether the brain tissue was homogenized in the standard volume or at a 50-fold higher dilution, indicating that PKC␣ was associated with PSD-95 and SAP102 in vivo before homogenization of the brain tissue (Fig. 7C). The novel finding that BAI1 and PKC␣ are associated with PSD-95 testifies to the utility of our proteomic approach in defining the potential pool of interaction partners for various PDZ domains.

DISCUSSION
Our results demonstrate proteins with COOH-terminal (E/ Q)(S/T)XV sequences possess the potential for stable interactions with the first two PDZ domains of PSD-95 and SAP102. This potential is further increased if P Ϫ1 is aspartate, glutamate, asparagine, or glutamine and if P Ϫ4 is a hydrophobic residue including leucine, isoleucine, valine, or perhaps tryptophan, lysine, or arginine (10). Aspartate and perhaps glutamate at P Ϫ10 may also foster binding to the first two PDZ domains of PSD-95 and SAP102. Nevertheless, the 0-, Ϫ2-, and Ϫ3-positions are highly sensitive to the precise residue present and are most crucial in determining the selectivity of the NR2b COOH terminus for binding to the first two PDZ domains of PSD-95 and SAP102. We found a severalfold preference of these PDZ domains for valine at P 0 over the related amino acids leucine and isoleucine, for threonine at P Ϫ2 over serine, and for glutamate and glutamine at P Ϫ3 over aspartate. Accordingly, even conservative substitutions at these positions such as serine for threonine or aspartate for glutamate affect binding. As discussed in the following paragraphs, these findings are in good agreement with published results on proteins that bind to these PDZ domains (see Table I).
The first identified binding partners for PSD-95 were the NR2 subunits and Kv1.4. They not only match the (E/Q)(S/ T)XV motif; they also possess an aspartate or glutamate at P Ϫ1 and an isoleucine, leucine, or valine at P Ϫ4 . They are therefore likely to be among the strongest interactors in vivo. Kv1.1, Kv1.2, and Kv1.3 also have (S/T)XV motifs and hydrophobic residues at P Ϫ4 and aspartate at P Ϫ1 . However, they possess hydrophobic residues at P Ϫ3 , which should strongly reduce binding to the first two PDZ domains of PSD-95 and SAP102 (Figs. [3][4][5]. In fact, PSD-95 association with Kv1.1, Kv1.2, or Kv1.3 appears to be much weaker than with Kv1.4 in a yeast two-hybrid assay and a clustering assay in mammalian cells (10,72). In a clustering assay, Shaker, a Drosophila homolog of the Kv1 subfamily, exhibits an interaction with PSD-95 that is as strong as the interaction between Kv1.4 and PSD-95 (72). In contrast to Kv1.1, Kv1.2, and Kv1.3, Shaker perfectly matches positions 0 through Ϫ3 of the Kv1.4 COOH terminus, and P Ϫ4 carries an isoleucine. These findings corroborate our consensus sequence for binding to the first two PDZ domains of PSD-95 and SAP102.
The next two proteins in Table I are the inwardly rectifying K ϩ channels Kir2.1 and Kir2.3 (24,51). Both proteins have an isoleucine rather than a valine at P 0 , which should strongly reduce binding affinity for those PDZ domains. However, they carry an arginine at P Ϫ4 ; an aspartate or glutamate, respectively, at P Ϫ10 ; and, in the case of Kir2.1, a glutamate at P Ϫ1 . According to our analysis, these residues help to strengthen the interaction with our PDZ domains and may promote stable binding even if the COOH-terminal position is not valine but the related isoleucine. Kir3.2 and Kir3.3 do conform to the generic (S/T)XV motif, but a careful analysis indicates that they do not bind to PSD-95 or SAP102 (24). The lack of detectable interaction is probably due to the presence of the positively charged lysine at P Ϫ1 , which inhibits binding to our NR2b-FIG. 7. Co-immunoprecipitation of PSD-95 and SAP102 with BAI1 and PKC␣ but not frizzled 2. Rat forebrain (0.8 g) was homogenized either in 8 ml (1ϫ in A and C) or 400 ml (B, 50ϫ in A and C) sucrose buffer before collection of a crude membrane fraction by ultracentrifugation and solubilization with deoxycholate. A, proteins were immunoprecipitated with anti-BAI1 or anti-frizzled 2 antibodies or with nonimmune control rabbit IgG. Immunoblotting for PSD-95 and SAP102 revealed that PSD-95, but not SAP102, is present in the same immunocomplex as BAI1. Immunoprecipitation of frizzled 2 did not result in co-immunoprecipitation of either of the PDZ domain-containing proteins. The Load lane represents the deoxycholate solubilized membrane fraction that was used for immunoprecipitation. B, immunoblot for frizzled 2 showing that frizzled 2 was effectively immunoprecipitated from the solubilized membrane fraction used in A. C, immunoprecipitations of PKC␣ and immunoblotting for PSD-95 and SAP102 were performed in a fashion similar to that in A. Mouse IgG was used as control. Both PSD-95 and SAP102 are found in the same immunocomplex as PKC␣. Each experiment was repeated at least three times with similar results. derived peptides at this position. Similarly, the ␤ 1 adrenergic receptor carries an ESXV motif with lysine at P Ϫ1 and does not appear to interact with either the first or the second PDZ domain of PSD-95 (28); it does bind to the third PDZ domain of PSD-95 (28), indicating that the third PDZ domain is quite different from the first two PDZ domains. The ␤ 2 adrenergic receptor with the COOH-terminal DSPL sequence binds to the PDZ domain of the Na ϩ /H ϩ exchange regulatory factor (53) but does not seem to associate with PSD-95 (28), probably because of the COOH-terminal leucine without compensatory residues at the Ϫ1-, Ϫ4-, or Ϫ10-position.
The plasma membrane Ca 2ϩ ATPase PMCA4b has a COOHterminal interaction sequence that is predicted to allow for high affinity interaction: threonine at P Ϫ2 , the hydrophobic leucine at P Ϫ4 , and aspartate at P Ϫ10 . It strongly binds to PDZ1 and -2 of PSD-95 and SAP102 (16,52). PMCA2b also binds to the first two PDZ domains of PSD-95, but these interactions are much weaker than those observed with PMCA4b (16), and no binding could be observed between PMCA2b and SAP102 (52). This binding behavior would be predicted from our findings that the leucine at the 0-position of PMCA2b should strongly reduce binding to PDZ1 and -2 of PSD-95 and SAP102. Maguin-1 matches the (E/Q)(S/T)XV motif, co-immunoprecipitates with PSD-95 from rat brain extracts, and binds to one or more PDZ domains of PSD-95 (73). Although the exact PDZ domains that interact with Maguin-1 have yet to be defined, based on our analysis it could bind to the first two domains because it has not only the (E/Q)(S/T)XV motif at its COOH terminus but also an isoleucine at P Ϫ4 . However, the histidine at P Ϫ1 is partially positively charged and may potentially be inhibitory for the binding to the first two PDZ domains and perhaps direct it to the third PDZ domain.
CRIPT also possesses that (E/Q)(S/T)XV motif and a lysine at P Ϫ4 , which, like isoleucine at this position, increases binding to the first two PDZ domains of PSD-95 and SAP102. Although it was originally described mainly as a binding partner for the third PDZ domain of PSD-95 (12), CRIPT did interact with the second PDZ domain in the original yeast two-hybrid assays (12). A peptide derived from the COOH terminus of CRIPT bound in our titration assays with K D values between 1 and 2 M to the second PDZ domains of PSD-95 and SAP102, although binding to the first PDZ domains was either very weak or not detectable (Fig. 2). We propose that the COOH terminus of CRIPT strongly interacts not only with the third but also the second PDZ domains of PSD-95 and SAP102. This binding pattern would explain why application of a membrane-permeable peptide derived from the CRIPT COOH terminus effectively dispersed PSD-95 from the postsynaptic site (74).
Similar to CRIPT's QTSV COOH-terminal sequence, Citron has the sequence QSSV at its COOH terminus and binds preferentially to the third and possibly also to the second PDZ domain of PSD-95 (15,75). It carries aspartate at P Ϫ4 , in contrast to CRIPT's lysine, and P Ϫ2 is a serine rather than a threonine; accordingly, we predict that Citron binds in vivo much more weakly than CRIPT to the second PDZ domain of PSD-95. Indeed, Citron's interaction with PDZ2 of PSD-95 appears to be much less favorable than with PDZ3 (15). The Citron-related COOH-terminal sequence EESSV is present in the semaphorin Sema4C and predicts a modest affinity for the first two PDZ domains of PSD-95 and SAP102. Sema4C coimmunoprecipitates with PSD-95 from rat brain and binds to a fusion protein containing the first two PDZ domains of PSD-95 (76). Another semaphorin, Sema4f, which has also recently been shown to interact with the second PDZ domain of PSD-95, possesses an advantageous threonine at P Ϫ2 but a disadvantageous isoleucine at P 0 (77).
SynGAP is a Ras-GTPase-activating protein initially identified in a subcellular fraction enriched with postsynaptic structures (13) and also by a yeast two-hybrid screen with the third PDZ domain of SAP102 (14). SynGAP does match the (E/Q)(S/ T)XV motif but does carry an arginine at P Ϫ1 . Because P Ϫ1 does not tolerate a lysine well, we would expect that an arginine at this position also inhibits binding to the first two PDZ domains of PSD-95 and SAP102. However, we have not tested this prediction, and it is possible that P Ϫ1 does accept an arginine. Yeast two-hybrid interaction assays have suggested that Syn-GAP binds not only to the third but also to the first two PDZ domains (14). However, the interaction between the first two PDZ domains and SynGAP was not confirmed by other tests. Because the yeast two-hybrid system does result in a substantial portion of false positive results, it is quite conceivable that SynGAP may interact with the third but not the first two PDZ domains of PSD-95 and SAP102 in intact neurons, as would be predicted by our studies.
Having an asparagine at P Ϫ3 , the neuregulin receptor ErbB4 does not conform to the (E/Q)(S/T)XV motif but does interact with the first two PDZ domains of PSD-95 and SAP102 (63,78). ErbB4 does have an arginine at P Ϫ4 , which, like lysine (Figs. [3][4][5], may foster binding to the first two PDZ domains of PSD-95 (10). We hypothesize that the conservative, although not ideal, substitution of a glutamate or glutamine at P Ϫ3 by an asparagine can be counterbalanced by arginine at P Ϫ4 , thereby permitting the PDZ domain interaction. Neuroligin is another protein that has a generic SXV motif but does not match the (E/Q)(S/T)XV sequence. It has a threonine at P Ϫ3 and an arginine at P Ϫ1 , which may exert a negative effect at this position. In agreement with our data, no binding to the first two PDZ domains of PSD-95 was detectable in earlier studies, although it does interact with the third PDZ domain of PSD-95 (30).
Finally, affinity chromatography with recombinant SAP102 protein identified one of the COOH-terminal splice variants of a novel protein, Nedasin, as a binding partner for the first two PDZ domains of SAP102 (79). Like ErbB4, Nedasin does not have glutamate or glutamine but rather has a serine at P Ϫ3 . It does have a hydrophobic residue at P Ϫ4 , which may in part compensate for the lack of the more optimal glutamate/glutamine at P Ϫ3 .
In summary, of the 20 proteins that have been shown to bind with a COOH-terminal (S/T)XV motif to the first two PDZ domains of PSD-95 or SAP102, 11 have an (E/Q)(S/T)XV sequence and should interact with these PDZ domains in a stable manner and with high affinity (Table I). The other nine binding partners have variations either at the 0-or Ϫ3-position that are predicted to weaken the interactions and at the same time residues at the Ϫ1or Ϫ4-position and in some cases also at the Ϫ10-position that strengthen the interactions and perhaps compensate for the variations at the 0-and Ϫ3-positions. Of these nine proteins, only a few have been shown to actually interact with PSD-95 and SAP102 by coimmunoprecipitation experiments from the brain. We conclude that there is a substantial amount of flexibility in the COOH-terminal binding sequences of ligands of the first two PDZ domains of PSD-95 and SAP102. Most proteins with the optimal (E/Q)(S/T)(D/E/Q/ N)V sequence have been shown or are likely to interact with those PDZ domains (see Table II and Fig. 7). Deviation from this sequence often results in weaker interactions. Many of the weaker interactions may have an auxiliary character by acting primarily in concert with other interactions within a protein complex, thereby contributing to the overall binding of two or more proteins in the context of multiple protein-protein interactions.
The third PDZ domains of PSD-95 and SAP102 possess bind-ing requirements that are quite different from the first two PDZ domains. Many of the proteins that bind to the first two PDZ domains do not effectively interact with the third, including NR2 subunits and the various Shaker-type and inwardly rectifying K ϩ channels (Table I). Furthermore, the wild-type NR2b COOH-terminal peptide as well as peptides based on the wild-type NR2b sequence with single amino acid residue substitutions (see Figs. 3 and 4) do not exhibit any specific binding to PDZ3 of PSD-95 and SAP102 (data not shown). Other proteins, including the ␤ 1 adrenergic receptor and Neuroligin, associate with the third rather than the first two PDZ domains of PSD-95 or SAP102 (Table I). A sequence alignment of the PDZ domains from PSD-95 and SAP102 (Table III, lower part) shows that PDZ1 and -2 have a substantially higher degree of homology among themselves than with the PDZ3 domains, although all of them are type I PDZ domains. This study focused on the interaction between NR2b and PDZ1 and -2 of PSD-95 and SAP102. The exact determinants of the specificity for the third PDZ domain could not be explored within this study, because it would require a completely different set of peptide libraries. However, it appears that no single amino acid substitution of the NR2b consensus sequence will allow binding to PDZ3, since none of the peptides listed in Fig. 3 and 4 showed specific binding to either of the third PDZ domains of PSD-95 or SAP102.
To evaluate our results with respect to available structural information, we have modeled the binding of an NR2b-based peptide in the second PDZ domain of PSD-95. Our approach is based on the NMR solution structure of PSD-95 PDZ2 (80) and the co-crystal structure of PSD-95 PDZ3 with a peptide based on the CRIPT COOH terminus (23). The peptide IESDV was modeled in the same ␤-strand conformation as the CRIPT peptide KQTSV (23) and manually docked into the binding pocket of PDZ2 using Sybyl (Tripos). Some visualization was also performed with WebLab Viewer Lite (Accelrys) (Fig. 8).
Stabilizing hydrogen-bonding interactions are expected based on this theoretical approach that closely match the interactions observed in the co-crystal structure of PDZ3 and CRIPT (23). Table III (upper section) summarizes residues in PDZ2 important for binding to NR2b, based on previous structural and these modeling data as well as our peptide binding assays.
The free carboxyl terminus of valine at P 0 is predicted to form hydrogen bonds with the carbonyl oxygens of Leu 170 , Gly 171 , and Phe 172 in the highly conserved GLGF backbone, in agreement with the PDZ3-CRIPT peptide structure (23) (Fig. 8 and data not shown; see Table III for an alignment of these residues with other PDZ domains). The valine hydrophobic side chain is oriented into a hydrophobic pocket lined with side chains from residues in the GLGF motif (Phe 172 and Ile 175 ) and ␣B helix (Val 229 and Leu 232 ; Ile 175 and Leu 232 are not depicted in Fig. 8, but see Ref. 23). Substitutions of other hydrophobic residues (alanine, leucine, and isoleucine) have a detrimental effect on binding to the second PDZ domain (Fig. 3). Perhaps the conformation of the residues that line the hydrophobic pocket is such that only valine at P 0 has the right side chain length that allows a good fit. Alanine would be too small to fill the volume, creating energetically unfavorable unfilled space. Although leucine and isoleucine can be accommodated in this position to some degree, their longer side chains might be too large to fit in the pocket as well as valine's side chain does, decreasing the overall affinity of the corresponding peptides for the PDZ domains.
Analogous to the Ϫ1-position serine side chain in the cocrystal of PDZ3 and the CRIPT peptide (23), the P Ϫ1 aspartate side chain of the NR2b peptide is oriented in the ␤-strand such that its carboxyl side chain points out of the PDZ binding pocket. In contrast to predictions based on the co-crystal structure for PDZ3-CRIPT peptide (23), the carboxyl oxygen of the aspartate residue side chain may participate in hydrogen bonding interactions with the hydroxyl side chain of Ser 173 (Fig. 8; see also Ref. 80) or the amine hydrogens of Lys 193 of the second PDZ domain (Fig. 8; although they may not form simultaneously, all three hydrogen bonds are indicated). It is possible that Lys 193 is not readily available for formation of a hydrogen bond with the aspartate at P Ϫ1 , because it may be tied up in a hydrogen bond with the glutamate or glutamine at P Ϫ3 (see below). The residue that corresponds to Ser 173 in PDZ2 is conserved in all PDZ1 and -2 domains of PSD-95 and of its homologs SAP97, SAP102, and PSD-93 and also in the NR2interacting PDZ5 domains of MAGI-1, -2, and -3 (Table III). However, this residue is an asparagine in PDZ3 of PSD-95 (Asn 326 ) and its homologs (Table III). When Asn 326 is mutated to a serine, a peptide derived from the COOH terminus of Kv1.4 that normally binds only to PDZ1 and -2 can now also bind to the third PDZ domain (12). The potential hydrogen bonding interactions with Ser 173 may explain the preference for residues with negatively charged, electron-donating side chains (aspartate, glutamate) at P Ϫ1 . Positively charged (lysine) or hydrophobic residues substituted at P Ϫ1 greatly reduced affinity for the PDZ domain.
The CRIPT/PDZ3 co-crystal structure predicted that the hydroxyl side chain of the threonine at P Ϫ2 forms hydrogen bonding interactions with the nitrogen of the imidazole side chain of His 372 , which is conserved in most type I PDZ domains (e.g. Table III). The Ϫ2-position of wild-type NR2b is a serine, which according to our modeling can also interact with the corresponding His 225 in PDZ2 (Fig. 8). In our binding studies, we observed that substituting the P Ϫ2 serine with a threonine results in an ϳ2-fold increase in affinity. One potential explanation for this result is that the hydrophobic Val 229 in PDZ2, which is one ␣-helical turn away from His 225 , lines the binding pocket of and is in close proximity to the residue at P Ϫ2 of the ligand (Fig. 8). The interaction of the NR2b S2T ligand in the binding pocket of the PDZ domain may be stabilized by hydrophobic interactions between the methyl group of the threonine side chain and the hydrophobic side chain of Val 229 . Because serine lacks this methyl group, the interactions of the wild-type NR2b peptide would be weaker. Val 229 is conserved in PDZ1 and -2 of PSD-95 and its homologs; it is isoleucine in PDZ5 of MAGI-1, -2, and -3 (Table III). However, this position carries an alanine in PDZ3 of the PSD-95 homologs (Table III), which may provide less if any hydrophobic interactions with the threonine side chain.
The carboxyl side chain of glutamate at P Ϫ3 can form a hydrogen bond with Thr 192 or with Lys 193 of PDZ2 ( Fig. 8; we show both potential hydrogen bonds, although only one of the two may exist at any given time). Our results indicate a substantially reduced affinity when this glutamate is replaced by aspartate but not when it is substituted by glutamine. Glutamine possesses the length and the potential to form hydrogen bonds with Thr 192 or with Lys 193 comparable with glutamate but does not carry a negative charge. Aspartate does have a negative charge, but it is shorter than glutamine. Our findings, therefore, indicate that the main factor for strong binding to PDZ2 of PSD-95 is the side chain length and the capability to form a hydrogen bond with either Thr 192 or Lys 193 . It does not appear to be the negative charge of the glutamate, although the latter may stabilize binding by electrostatic interactions with Lys 193 . Such considerations may especially be true for ligand binding to PDZ1 of PSD-95, in which case changing glutamate to a glutamine at P Ϫ3 results in a substantial reduction in affinity (Fig. 3). Thr 192 and Lys 193 are conserved in all PDZ1 and -2 domains of the PSD-95 homologs (Table III). However, the PDZ5 domains of MAGI-1, -2, and -3 have leucine and arginine at the positions homologous to Thr 192 and Lys 193 , respectively, and all PDZ3 domains of PSD-95 homologs have serine (Ser 339 ) and phenylalanine at the corresponding positions (Table III). The glutamine residue at P Ϫ3 of the PDZ3 ligand CRIPT can form a hydrogen bridge with Ser 339 and also with Asn 326 (23). However, the high affinity PDZ3 ligand Neuroligin has threonine rather than glutamine or glutamate at P Ϫ3 , indicating that PDZ3 of PSD-95 and its homologs have requirements different from those of PDZ1 or -2 for strong binding.
The Ϫ4-position of the ligand is at the border of the PDZ binding pocket proper. The CRIPT/PDZ3 co-crystal was unable to resolve the structure at and beyond this position in the CRIPT peptide (23). However, the NMR structure of the ␣-syntrophin PDZ domain in complex with a peptide derived from the COOH terminus of syntrophin's binding partner Na v 1.4 (KESLV) indicates that Asp 143 forms a salt bridge with lysine at P Ϫ4 (81) (Asp 143 corresponds to Glu 226 in PSD-95 PDZ2; FIG. 8. Structural model of the ligand peptide IESDV with the second PDZ domain of PSD-95. Modeling was based on the crystal structure of the PDZ3 PSD-95 obtained with a peptide from CRIPT (Protein Data Bank accession number 1BE9) and the average NMR solution structure of PSD-95 PDZ2 (Protein Data Bank accession number 1QLC). The PDZ2 was aligned visually with PDZ3 using Sybyl (available on the World Wide Web at www.tripos.com), and amino acid residues of CRIPT were changed into the COOH-terminal sequence of NR2b. Visualizations were rendered with WebLab Viewer Lite 4.0 (available on the World Wide Web at www.accelrys.com). The peptide is represented as a ball and stick. The PDZ domain is rendered with a Connolly solvent surface, colored according to electrostatic potential (blue, basic; red, acidic). Residues in the PDZ2 domain that are potentially important for interaction with the peptide have been revealed. Several important potential hydrogen bonding interactions are depicted as dashed green lines. Table III). Mutating Asp 143 to glycine nearly abolished the binding of this peptide to syntrophin's PDZ domain (25). Our interaction assays with PDZ2 of PSD-95 and SAP102 indicate a preference for either hydrophobic or positively charged residues at P Ϫ4 . Aspartate is not tolerated at this position. The situation appears to be different for PSD-95-PDZ1, which shows reduced affinity for the lysine-substituted peptide, and for SAP102-PDZ1, which tolerates an acidic aspartate substitution at this position in our fluorescence polarization assays (Fig. 3). In fact, ␣-syntrophin's Asp 143 is substituted with a glutamate in all PDZ2 domains and with a serine in all PDZ1 domains of PSD-95 and its homologs. We therefore propose that lysine at P Ϫ4 may favorably interact with Glu 226 of the PSD-95 PDZ2 domain and the corresponding residues in the other PDZ2 domains. If there is a hydrophobic residue at P Ϫ4 as in the WT NR2b sequence, the peptide may adopt a different conformation so that the hydrophobic side chain is located in a more hydrophobic environment as shown in Fig. 8. Substitutions further upstream from the Ϫ4-position generally did not have as dramatic effects on binding as did substitutions closer to the COOH terminus, as would be expected, although it is also clear that binding to PDZ domains is influenced by positions more NH 2 -terminal than the last 4 residues.